Multi-Source Precipitation Data Fusion Across Continental United States

Dataset Description: This dataset supports our research presented in the paper "A deep learning-based framework for multi-source precipitation fusion" by Gavahi et al., (2023) published in Remote Sensing of Environment. The study introduces a novel deep learning architecture for merging and downscaling multiple precipitation products, aiming to enhance quantitative precipitation estimation (QPE) accuracy. The developed model, the Precipitation Data Fusion Network (PDFN), integrates 3D-CNN and ConvLSTM layers to capture the inherent spatiotemporal dependencies of precipitation data. The results indicate significant improvements in error statistics. The dataset includes merged daily precipitation estimations using the PDFN model. The data cover the Continental United States (CONUS) and are provided at a spatial resolution of 0.05 degrees. Temporal coverage spans from January 1, 2015, to April 30, 2024. The coordinate reference system used is WGS1984. Note: Since the PERSIANN-CDR dataset is only available until the end of 2023, in this dataset, we used PDIR-Now instead to ensure the dataset's continuity and completeness. In the original paper, we used PERSIANN-CDR, but here we used PDIR-Now to extend the dataset to cover the period until April 30, 2024. Usage Notes: This dataset is intended for use in applications such as land surface modeling, flood forecasting, drought monitoring and prediction. Users are requested to cite the associated paper when utilizing the dataset for academic or research purposes. Related Publications: For further details on the methodology and applications of this dataset, refer to the paper "Gavahi, K., E. Foroumandi, and H. Moradkhani (2023), A deep learning-based framework for multi-source precipitation fusion, Remote Sensing of Environment, doi:10.1016/j.rse.2023.113723"数据集说明：本数据集支撑了Gavahi等人2023年发表于《Remote Sensing of Environment》的论文《基于深度学习的多源降水融合框架》（A deep learning-based framework for multi-source precipitation fusion）中的相关研究。该研究提出了一种新颖的深度学习架构，用于多降水产品的融合与降尺度，旨在提升定量降水估算（Quantitative Precipitation Estimation, QPE）的精度。所开发的降水数据融合网络（Precipitation Data Fusion Network, PDFN）模型，集成了三维卷积神经网络（3D-CNN）与卷积长短期记忆网络（ConvLSTM）层，以捕捉降水数据固有的时空依赖性。研究结果表明，该模型的误差统计指标得到了显著改善。 本数据集包含基于PDFN模型生成的逐日融合降水估算产品。数据覆盖美国本土（Continental United States, CONUS），空间分辨率为0.05度，时间跨度为2015年1月1日至2024年4月30日，采用的坐标参考系为WGS1984。 注意：由于PERSIANN-CDR数据集仅可获取至2023年底，本数据集改用PDIR-Now以确保数据的连续性与完整性。原论文中使用的是PERSIANN-CDR，本次通过替换为PDIR-Now将数据集的时间覆盖范围延伸至2024年4月30日。 使用说明：本数据集可应用于陆面模拟、洪水预报、干旱监测与预警等场景。若将本数据集用于学术或研究用途，请引用相关论文。 相关文献：如需了解本数据集的方法学与应用细节，请参阅论文：Gavahi, K., E. Foroumandi, 和 H. Moradkhani (2023), 基于深度学习的多源降水融合框架, 《Remote Sensing of Environment》, doi:10.1016/j.rse.2023.113723